Regression in Spec::BeAExpectation(T)

[Sija]

Merging #5646 caused a regression:

require "spec"

describe "Spec::BeAExpectation(T)" do
  it { Int32.should be_a Int32.class }
end

v0.24.1 (carc.in):

.

Finished in 83 microseconds
1 examples, 0 failures, 0 errors, 0 pending

master:

Error in line 1: instantiating 'be_a(Class)'

in /.../crystal/src/spec/expectations.cr:271: can't use Object as generic type argument yet, use a more specific type

      Spec::BeAExpectation(T).new

[luislavena]

I always assumed be_a refers to instances of instead of kind of classes.

If not mistaken, RSpec does the same distinction between be_kind_of and be_instance_of:

• https://relishapp.com/rspec/rspec-expectations/docs/built-in-matchers/type-matchers
• http://www.rubydoc.info/gems/rspec-expectations/RSpec/Matchers/BuiltIn/BeAKindOf
• http://www.rubydoc.info/gems/rspec-expectations/RSpec/Matchers/BuiltIn/BeAnInstanceOf

Perhaps the same approach needs to be made here in order to properly differentiate kind from instances.

Cheers.

[Sija]

IIRC that's what be should do, be_a is/was meant to check for given type, see previous definition:

crystal/src/spec/expectations.cr

Lines 270 to 272 in 844cc6e

	macro be_a(type)
	Spec::BeAExpectation({{type}}).new
	end

Now, it reads:

crystal/src/spec/expectations.cr

Lines 270 to 272 in ed0aad8

	def be_a(type : T.class) forall T
	Spec::BeAExpectation(T).new
	end

type argument is still there, yet in actual version it seems to work differently.

[luislavena]

@asterite any comment on this?

[Sija]

@asterite ping!

[asterite]

My commit should be reverted

[RX14]

(just the be_a bit)